Вы находитесь на странице: 1из 6

2011 American Control Conference on O'Farrell Street, San Francisco, CA, USA June 29 - July 01, 2011

978-1-4577-0079-8/11/$26.00 2011 AACC

4598

and nally designed a closed loop control in [5]. The performance of combustion engine test benches can for example be signicantly improved by model reference adaptive control (see e. g. [6]), robust multivariable feedback control (see e. g. [7]) and robust inverse control (see e. g. [1]). Controlling the hydrodynamic brake itself has rarely been treated in the past. Common implementations at test benches make use of simple feedback controllers extended by some maps to compensate deviations. The design of a controller for the torque of a water brake is described in this work. An approximative inversion of the plant is generated by means of a feedforward control, while a feedback controller compensates uncertainties and disturbances. Only one simple feedback controller needs to be tuned after measuring the nonlinear static maps. The developed controller has been tested on a test bench with a hydrodynamic dynamometer with a maximum power of 420 kW used to load a heavy duty truck engine with an output of approximately 250 kW running at a constant speed of 1000 rpm. The paper is organized as follows: A brief description on the working principle and the estimation of a Wiener type model of a water brake dynamometer is given in Section II. The following section deals with the developement of the controller for constant speed. Finally, a comparison is made between the developed controller and a standard implementation. Conclusions including future aspects will complete the paper. II. H YDRODYNAMIC DYNAMOMETERS This section deals with the working principle of a water brake and the developement of a simple model based on real measurements. It turns out, that the behaviour can be described by means of a Wiener type model. A. Working principle A hydrodynamic brake consists of two essential parts (see Fig. 3): The rotor driven by the device under test, composed by the turbine wheel and the shaft, and the stator unit composed of the housing and the supply. A water brake operates on the F ottinger principle (see [8]). In contrast to torque converters or clutches this type of dynamometer has only one rotating part. As a consequence the slip calculated by the relative difference of the rotor and stator speed is always 100%. This results in a large thermal impact on the working uid making a uid exchange indispensable. The energy dissipation process occurs entirely in the working chambers. From entering the water brake through the inlet valve until leaving it by the outlet valve, the uid passes the gap between rotor and stator unit several times. Most of the thermal impact on the working uid results from this incidence losses. Furthermore, friction and secondary circulation losses lead to a warming of the uid. An electric machine offers sometimes the possibility to feed the braking energy back to the net. The appropriate efciencies of the machine, the frequency converter, etc. have

Stator VortexFlow

Rotor AirVent

WaterInlet

Fig. 3.

Section drawing of a hydrodynamic brake

to be taken into account. However, the electrical energy is transformed by resistors into heat in many cases. Using a water brake the whole mechanical energy transmitted by the connection shaft is converted into heat. A detailed overview of the energy dissipation as well as on the structural design of a hydrodynamic brake is given in [2] and [9] respectively. B. Modeling of a hydrodynamic dynamometer Fig. 4 shows the stationary dynamometer torque TD as a function of the inlet i and the outlet valve position o for a constant speed of n = 1000 rpm. The average of the measured torque over several seconds has been calculated after reaching the operating point to achieve accurate results even in the presence of noise. A valve position of 100% of both valves corresponds to the maximum possible torque. In this case the inlet valve is fully opened and the outlet valve almost closed. Especially for lower speeds a signicant leakage ow can be determined for these and similar valve positions. Fig. 4 and Fig. 8 indicate a band of high gradients in the dynamometer torque TD , while in other regions the torque TD is saturated (see Fig. 4). A low ll level according to a nearly closed inlet valve while the outlet valve is wide opened will in any case give a low dynamometer torque TD . The difference between the actual torque and the maximum possible torque at a wide opened inlet valve and a nearly closed outlet valve is fairly limited. Similar maps can also be recorded for other speeds. However, for higher speeds the steep rise of the dynamometer torque TD is shifted to higher values of both valve positions. The insensitive area next to the origin becomes bigger. In this case also the maximum torque is higher.

4599

Speed n = 1000 rpm


D

800 Dynamometer torque T in Nm

600

400

200

0 100 80 60 40 20 Inlet valve position in %


i

100 80 60 40 20 0 0 Outlet valve position o in %

of the ll level occurs within a shorter time. Therefore, the response time of the water brake to changes of the operational point decreases with higher speed. In a rst attempt, an estimation of the linear dynamics was performed to determine the worst case scenario. For this scenario the controller has been developed and tuned in simulation. As the characteristic of the water brake changes with the inlet i and the outlet valve position o as well as the speed nD , a further extension of the linear dynamics in form of LPV contributes to an improvement. For more information on LPV modelling, identication and control see e. g. [10], [11] and [12]. Above observations suggest the usage of a Wiener type model (see Fig. 6) where the linear dynamics is downstreamed by a nonlinear static map (). It is x = A x (t ) + B u (t ) z = C x (t ) y (t ) = (z) (1) (2) (3)

Fig. 4. Dynamometer torque TD as a function of the valve positions for a constant speed

To increase the performance of the subsequent developed inverse torque control and to obtain more realistic simulation results, it is necessary to combine the above described maps with dynamics. Fig. 5 shows the response of the hydrodynamic dynamometer to steps of the inlet valve position i for different speeds nD .
Outlet valve position o = 50 % Inlet valve position i in % 70 60 50 40 30 0

with x representing the state of the linear dynamics, z the output of the linear dynamics and y the output of the Wiener type model.

Linear dynamics

Nonlinear static map ()

10

12

14

16

18

20

Fig. 6.

Wiener type model

Dynamometer torque TD in Nm

500 400 300 200 100 0 0 2 4 6 8 10 Time in s 12 14 16 18 20 Dynamometer speed nD = 500 rpm Dynamometer speed nD = 1000 rpm Dynamometer speed nD = 1500 rpm

The input u consists of the inlet valve position i , the outlet valve position o and the speed nD . The output y is given by the dynamometer torque TD . Common identication procedures for Wiener type models are based on two step approaches, see for example [13], [14] and [15]. III. I NVERSE T ORQUE C ONTROL Fig. 7 shows the scheme of the inverse torque control. On the basis of the nonlinear static map () the set values for both valve positions are calculated from the desired dynamometer torque TD, set . A feedback controller is used to compensate uncertainties and disturbance effects. These effects may include model-plant-mismatch as well as disturbances to the hydrodynamic dynamometer caused by an operation point change of the combustion engine. The determination of the various parts of the controller will be shown for a constant speed below. The extension to variable combustion engine test bench speeds will be discussed in a later paper.

Fig. 5. Response of hydrodynamic dynamometer to steps of the inlet valve position i

The water brake including the valves and the actuators show a low pass behaviour. This behaviour is not only related to steps of the inlet valve position i but also caused by steps of the outlet valve position o and the dynamometer speed nD . However, the low pass characteristics shown in Fig. 5 changes with the dynamometer speed nD . Due to an increased centrifugal force at higher dynamometer speed nD , a change

4600

Speed n = 1000 rpm

i TD, set Feed forward controller o + +

i o
Hydrodynamic dynamometer TD

100
40 0

0 Dynamometer torque TD in Nm

75

90
20 0

30 0

600

10 0

80 70

50 0
70 0

Feedback controller

nD

Inlet valve position in %

50

40

60 50 40
10

10

30

20 0

60
50

0
700

50
10 0

30 20 10
10

20
50

30

40 0
0

Fig. 7.

Control scheme of inverse torque control

600 500

10 0

A. Inversion of the nonlinear static map As shown in Fig. 4 and Fig. 8, a specic torque can be achieved with an innite number of combinations of the inlet i and the outlet valve position o . Thus, this system offers an additional degree of freedom for control of the dynamometer torque TD . The introduction of an additional condition is required to obtain a unique control input (i , o ) achieving a given criterion of optimality. This condition can either be based on measurement quantities, on constraints imposed during the design of the controller or on a combination of both. The design goals for the controller contain both the tracking of the dynamometer torque TD as well as constraints of the temperature and the ow rate. For example, the maximum temperature at the outlet as well as the maximum temperature difference between inlet and outlet is specied by the manufacturer of the brake. Additional conditions may include the least possible movement of one or of both valves compared with the actual conguration or the least possible difference between both valve positions. A further condition could be the operation of both valves closer to the center of the operational area. In simulation as well as on the combustion engine test bench the following condition was implemented min
x

10

20

30

0 0

10

20

30

40 50 60 70 Outlet valve position in %


o

80

90

100

Fig. 8. Contour plot of dynamometer torque TD and vector eld representing the gradient

The gradient of the dynamometer torque TD for a constant speed nD = 1000 rpm is shown in Fig. 8. In the center of the operational area of both valves the gradient is not maximum, but reaches a high value and offers a trade-off between susceptibility to disturbances and sensitivity. On the other hand some freedom in the control of the dynamometer torque TD remains due to the design parameter x0 . For instance, x0 allows to consider the temperature of the working uid at the outlet in the control loop. B. Robustifying feedback control Neglecting the linear dynamics in Fig. 5 and disturbance effects, it is possible to control the plant solely by means of the above described inversion of the nonlinear static map. However, to take these phenomena into account a feedback controller is introduced subsequently. The calculation of two control variables based on one error signal is described with reference to Fig. 8. All possible combinations of inlet i and outlet valve position o that lead to one and the same dynamometer torque TD lie on a continuous curve. Partial deviations from this property can only be observed for very high and very low values for a given speed and the marginal areas. Notice also that these lines are almost parallel especially in the area with the largest gradient and tilted by about 45 to the axes of abscissae in Fig. 8. Thus, the shortest connection between two of these lines is given by the orthogonal direction under +45 to the axes of abscissae. The feedback controller exploits the previous consideration. If the achieved dynamometer torque TD is less than the desired one TD, set , certain values are added to both value o ) calculated by the inversion of nonlinear i , positions ( static map (). The situation is reversed for a negative error.

x x0 TD x x

(4)

s. t.

= TD, set x x

with p N and x = i o . x = i o and x = i o respectively describe some boundaries. x and x allow on the one hand an avoidance of specic operational areas as those specied by the manufacturer. A certain operation range for the feedback controller persists on the other hand. The argument in (4) describes the difference between the actual valve positions x = i o and the design point x0 = i, 0 o, 0 . In the implemention p is set to 2 corresponding to the usage of the euclidean norm. Thus, the valves are operated near to the center of their working range offering large movement in both directions.

4601

Dynamometer torque T

It is i + i i = o + o o = with i = C(TD, set TD )i (s) e o = C(TD, set TD )o (s) e. (7) (5) (6)

600 in Nm 400 200 0 0 50 100 150

Speed nD in rpm

(8)

1050 1000 950 900 850 0 Reference 50 Standard control 100 Inverse torque control 150

Outlet temperature in C

While one valve is further closed, the other one is further opened. The ll level of the hydrodynamic dynamometer and therefore the torque TD can be changed faster compared to an approach using only one valve. The choice of C(TD, set TD )i (s) = C(TD, set TD )o (s) is a further simplication. If the movement of one of the two valves caused by rejecting errors in the inversion of the nonlinear static map () or disturbance effects should for example be limited, different choices of the controllers are possible. Using only one valve is the extremum. The following simple structure of the feedback controller has been chosen for measurements on the test bench: kI (9) C (s) = kP (TD, set TD ) + (TD, set TD ) s with kP = 0.18 and kI = 0.07. Especially fast changes of the operating point result in control variables slightly smaller or larger than the (soft) boundaries specied in the inversion of the nonlinear static map (). Under infrequent circumstances the controller would require control variables smaller than 0% or greater than 100%. In this case, problems can be resolved by introducing an anti-windup-loop. IV. M EASUREMENTS AT THE C OMBUSTION E NGINE T EST B ENCH The objective of an engine test bench is to operate an internal combustion engine like in a car or a heavy duty truck. This objective is equivalent to the tracking of a torque and a speed prole at the crank shaft of the engine. The engine to be tested as well as the dynamometer are the actuators. In the present case the speed of the test bench is controlled with the standard controller acting on the combustion engine. The torque at the crank shaft is controlled via the dynamometer. The results when using the standard controller for the torque and the developed control structure are compared in the following: The desired TD, set and the measured dynamometer torque TD using both control concepts are shown in the rst plot of Fig. 9. The effects of controlling the dynamometer torque TD on the speed of the test bench are depicted in the second plot. Furthermore, a comparison is given between the valve positions and the temperature of the working uid at the outlet. The tracking of the reference torque can signicantly be improved by using the proposed controller. The rise time is shortened by an rapid movement of both valves. In case of an emptied brake, this also leads to a reduction of the response time.

Outlet valve position

80 60 40 20

in %

50

100

150

Inlet valve position

80 60 40 20

in %

50

100

150

30 25 20 15 10 0 50 Time in s 100 150

Fig. 9. Comparison between developed controller and standard implementation Trial 1

While over- and undershootings can be determined using the standard control, the developed control shows this behaviour hardly. There is only a short, limited undershoot after a sudden and large reduction of the desired dynamometer torque TD, set . Furthermore, it should be noticed that the temperature at the outlet is less using the proposed control. More water ows through the brake according to a more opened inlet and outlet valve. However, an adjustment is possible by a shift of the design point. A simple comparison between the results suggests an increased performance using the controller developed in this paper. However, the control of a combustion engine test bench is always a trade-off between the two outputs. A quantitive comparison using the weighting functions JTD = 1 N TD TD, set N k TD, set =1
2

(10)

4602

and JnD = 1 N nD nD, set N k nD, set =1


2

(11)

is done and summerized in a normalized fashion in Table I. N characterizes the total number of measurement points.
TABLE I C OMPARISON BETWEEN S TANDARD C ONTROL AND I NVERSE T ORQUE C ONTROL Costs JTD JnD 100 100 25.4 98.6

possible by using dynamics depending on the operating point in the control design instead of the worst case scenario. The employment of other structures for the feedback control and the proof of robustness could also be of interest. However, the next step in the overall process is given by the development of a control for the speed of the test bench. Subsequently, as described in [1], the couplings between internal combustion engine and dynamometer will be considered by using a multi-input multi-output model in control design. VI. ACKNOWLEDGMENTS The authors gratefully acknowledge the sponsoring of this work by the COMET K2 Center Austrian Center of Competence in Mechatronics (ACCM). R EFERENCES

Standard control Inverse torque control

Using the developed inverse torque control leads to nearly the same values of the weighting function of the test bench speed, while the improvement in tracking the torque is signicant. However, the parametrization of the feedback controller affects the coupling from torque control to test bench speed. Through a more smooth setting of the control the impact is reduced, but this also decreases the torque tracking performance. Other test runs allow the same conclusions. The overall performance of the system is greatly increased. V. C ONCLUSION AND O UTLOOK Measurements on a real engine test bench indicate that a water brake can be described by a Wiener type model. This paper describes the design of a controller for the torque of such a plant. The feedforward controller calculates the control signals for inlet and outlet valve according to the inversion of the nonlinear static map () of the Wiener type model. A feedback controller is used to compensate modelplant-mismatch, errors in the inversion of the nonlinear static map () and disturbance effects. Compared with available standard implementations operating the developed controller leads to an increased performance. The usage of a hydrodynamic brake is not limited to stationary measurements. The presented novel approach offers the opportunity to run dynamic tests like the Heavy-Duty FTP Transient Cycle (see [16]) on test benches equipped with hydrodynamic dynamometers. To release the limitation to xed speeds a map approximating the dynamometer torque TD as a function of the valve positions and the actual speed of the test bench nD is calculated by interpolation between the recorded nonlinear static maps (). First results both in simulation as well as on the test bench show a promising behavior. To handle all the above mentioned design goals the proposed Wiener type model has to be extended to the temperature at the outlet (or the difference temperature between inlet and outlet). Improved results should also be

[1] E. Gruenbacher and L. del Re, Robust inverse control for combustion engine test benches, in 2008 American Control Conference, June 11 13 2008. Seattle, Washington, USA. [2] M. Vetr, T. E. Passenbrunner, H. Trogmann, P. Ortner, H. Kokal, M. Schmidt, and M. Paulweber, Control oriented modeling of a water brake dynamometer, in 2010 IEEE Multi-Conference on Systems and Control, September 810 2010. Yokohama, Japan. [3] J. K. Raine and P. G. Hodgson, Computer simulation of a variable ll hydraulic dynamometer. Part 1: torque absorption theory and the inuence of working compartment geometry on performance, Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, vol. 205, no. 33, pp. 155 163, 1991. [4] P. G. Hodgson and J. K. Raine, Computer simulation of a variable ll hydraulic dynamometer. Part 2: steady state and dynamic open-loop performance, Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, vol. 206, no. 13, pp. 4956, 1992. [5] P. G. Hodgson and J. K. Raine, Computer simulation of a variable ll hydraulic dynamometer. Part 3: closed-loop performance, Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, vol. 206, no. 53, pp. 327336, 1992. [6] D. Yanakiev, Adaptive control of diesel engine-dynamometer systems, in 37th IEEE Conference on Decision and Control, 1998. Tampa, Florida, USA. [7] B. J. Bunker, M. A. Franchek, and B. E. Thomason, Robust multivariable control of an engine-dynamometer system, IEEE Transactions on Control Systems Technology, vol. 5, no. 2, pp. 189199, 1997. [8] B. Schlecht, Maschinenelemente 2. Pearson Studium, 1 ed., 11 2009. [9] H.-P. Dohmen, Entwicklung eines dynamischen Motorenpr ufstands in Tandem-Anordnung mit digitaler Regelung. PhD thesis, Fachbereich Maschinenbau, Technische Hochschule Darmstadt, Germany, 06 1993. [10] J. S. Shamma, Analysis and design of gain scheduled control systems. PhD thesis, Laboratory for Information and Decision Systems, Massachusetts Institute of Technology, USA, 05 1988. [11] J. S. Shamma, Linearization and gain scheduling, in The control handbook (W. S. Levine, ed.), CRC Press, 1996. [12] B. Bamieh and L. Giare, Identication of linear parameter varying models, International Journal of Robust and Nonlinear Control, no. 12, pp. 841853, 2002. [13] W. B. Er, An optimal two stage identication algorithm for hammerstein wiener nonlinear systems, Automatica, vol. 34, no. 3, pp. 333 338, 1998. [14] A. Hagenblad, Aspects of the Identication of Wiener Models. PhD thesis, Department of Electrical Engineering, Link oping Universitet, Sweden, 1999. [15] Q. Zhang, A. Iouditski, and L. Ljung, Identication of wiener systems with montonous nonlinearity, in 14th IFAC Symposium on System Identication, 2007. Newcastle, Australia. [16] E. Inc., Emission test cycles: Heavy duty ftp transient cycle, 1999. http://www.dieselnet.com/standards/cycles/ftp trans.html.

4603

Вам также может понравиться